Chicago (IL) – I am scratching my head over this one. Nvidia got a $200 million scar this year, caused by faulty GPUs with cracking solder bump material. The company quickly switched to a new material and promised that has fixed the problem. Charlie Demerjian from the Inquirer was curious if Nvidia was truthful, bought a shiny new Macbook Pro, cut it in half and had the GPU x-rayed. There it was again, the bad bump material that can cause the GPU to die. The evidence seems to be crushing. What is Nvidia going to tell users – and Apple?
If you recall the events from July and August of this year, there was a massive recall of notebooks with Nvidia GPUs from Dell, HP and Apple - and there is at least one more large manufacturer that has not come officially forward yet, according to sources. While Nvidia never explained what the source of the problem was, industry sources told us that Nvidia’s use of high-lead in solder bumps (tiny bumps that glue the chip to the green plastic base) was the cause: Under extreme temperatures as well as expanding and contracting, the material may be exposed to fatigue cracking, resulting in a failing GPU and an essentially useless computer until the GPU is replaced.
Nvidia switched to so called eutectic solder bumps with less lead and more tin. The problem with such a switch usually is that it isn’t simple and requires a substantial engineering effort to reroute and trim current levels. Nvidia repeatedly said that the problem in those GPUs was minor and was corrected. And realistically, not all GPUs with high-lead solder bumps were failing, but they all carried the potential of failing. Over the following weeks, we received numerous reports from Nvidia switching the material and then switching back to high lead and apparently there was even substrate issue from one particular for a certain period of time – an issue, which, if true, may haunt Nvidia in 2009 with additional recall claims.
The result of all of this is that Nvidia’s general credibility has suffered and it was clear that there would be questions about those new and widely praised GPUs in Apple’s new Macbook Pros. Nvidia said there were no issues. Apple, as usual, does not reply to inquiry emails. So, how do you know how to be sure? Charlie Demerjian from the Inquirer wanted to be absolutely certain. He bought a brand new 15” Macbook Pro with a 9400M chipset and a 9600 GT GPU in California and sent it to a lab, where the GPU was sawed in half and the solder bumps were examined under an electron microscope with an X-ray microanalysis system.
There is plenty of image material to prove Demerjian’s claim – that there are possibly critical high-lead solder bumps. It turned out that the 9400M uses the corrected eutectic solders and the 9600 GT the “bad” high-lead solders. Of course, that does not mean that these GPUs will fail, but it means that Nvidia continues to use what it in indicated it does not use. Demerjian shared some email conversation he had with Nvidia, in which an Nvidia vice president clearly claims “that the combination of material underfill and bump is different from the combination that was exhibiting the bump crack field failures earlier in the year” – which seems to be a questionable statement, if the results found by Demerjian are correct. The Nvidia representative noted that there is no evidence that the material in the 9600 GT is “bad”.
In fact, no one can say if or when those solder bumps will crack. What remains is the question about why Nvidia isn’t exactly forthcoming about this issue. It would certainly be helpful if Nvidia laid out all of its cards and explained to consumers what the real issue was and how it was corrected. We would also be interested how Apple will react to this new question created by Demerjian. However, it is rather unlikely that Apple will comment on this question publicly.